Why are so many people talking about Claude? Meet the most dangerous AI in the world Anthropic's Claude Mythos 5 artificial intelligence model, considered one of the most dangerous AI models in the world, created fake identities to try to convince a developer to approve malicious changes to an open source project during a cybersecurity test. The case was released by the AI ??Security Institute and is the latest in a series of incidents involving the most advanced AI systems. According to the institute, the behavior was identified during routine assessments to measure the models' ability to execute cyber attacks. ?? Do you have any reporting suggestions? Send to g1 In total, the researchers recorded 17 actions from Mythos 5 and another two from GPT-5.6-Sol, from OpenAI, related to attempts to circumvent security mechanisms using cyber classifiers. The episode reinforces experts' concerns that increasingly sophisticated AI models are developing the ability to combine technical skills with people manipulation strategies, known as social engineering. Model lights up warning Claude Mythos is considered by experts to be the most dangerous AI model in the world today. The system is described as extremely fast and capable of finding security holes that would go unnoticed by humans. Anthropic's Claude chatbot is among the most popular, rivaling ChatGPT and Gemini Reuters via BBC "If Mythos ends up in the wrong hands, companies could be paralyzed overnight," Izabela Anholett, AI specialist and founder of Nura AI, explained to g1. According to her, the risk lies in the model's ability to identify and exploit vulnerabilities in digital systems. "Mythos Preview [test version] has already found thousands of high-severity vulnerabilities, including some in all major operating systems and web browsers," Anthropic said. "Due to its high power and the risks it poses to national and general security, it was not released to the common public. Instead, Anthropic created a consortium of companies and government agencies with access to the tool", explains Diogo Cortiz. The group's members include Amazon, Apple, Microsoft, Google, Nvidia, Broadcom and Mozilla. Cortiz recalls that an example of the model's effectiveness came from a Mozilla report. By using Mythos to analyze the Firefox browser, the organization identified a large number of previously unknown vulnerabilities and flaws. According to Anthropic itself, preliminary versions found thousands of high-severity vulnerabilities in operating systems and browsers, which is why access was restricted to a group of companies and government bodies. This year, a version derived from the model, called Fable 5, had its distribution temporarily suspended by the United States government due to concerns related to national security before being released again. Understand in the article below: New episode expands concerns At the end of July, Anthropic reported that a configuration error gave Mythos access to the internet during cybersecurity assessments, allowing the model to access the systems of three real companies. OpenAI revealed that one of its experimental models managed to exploit vulnerabilities and compromise a test server after receiving access to external tools. Although all episodes occurred in test environments, they raised concerns about the potential of these systems to carry out increasingly complex digital attacks. READ ALSO: OpenAI discovers more cases where its AI escaped isolation, says agency